Malaysian Ad Angles by Vertical (2026)

Across every running Malaysian ad, showcase leads at 70%, discount or offer follows at 44.6% and feature callout at 40.5%. Social proof is 18.1% and testimonial 12.5%. That ranking is the same in almost every vertical, which is exactly why publishing it as advice would be useless. If your category leads on discounts and so does everyone else's, you have learned nothing about your category. The reason these do not add to anything is worth stating plainly. An ad can carry several angles at once, and most do, so each figure is the share of ads carrying that label rather than a slice of a whole. Anyone turning this into a pie chart is drawing a picture of something that does not exist. Coverage is close to total. 222,656 of 222,921 running Malaysian ads carry at least one angle label, and 222,608 of those carry one this study reports on. The gap, 48 ads, is the tail of labels the model emits that our controlled vocabulary does not name, and it is published as a size rather than glossed over.

The index is the whole page. It divides a vertical's share of an angle by the national share of the same angle, so one is the market rate and two means twice as likely. Beauty and personal care runs a before-and-after angle on 15.8% of its ads against 6.3% nationally, an index of 2.51. Healthcare runs testimonial on 26.8% against 12.5%, an index of 2.14. Education runs case study on 6.8% against 1.7%, an index of 4. Food and beverage over-indexes on founder story at 2.14. None of that is a recommendation. Prevalence is not performance, and this archive holds creative metadata with no spend attached, so a vertical that leans on an angle may be leaning on it because it works or because everyone in it copies everyone else. What the index gives you is a map of where your category's creative already crowds, which is a different and more useful thing than a list of best practices. The software and finance rows carry the largest indices on the page and the smallest practical meaning: both over-index on an interface angle, which is what a screenshot of a product is, and almost nothing outside those two verticals is a screenshot.

Nationally, video is 48.5% of running Malaysian ads and static image 30.3%. Beneath them the carousel variants and Meta's dynamic formats divide the rest. Format separates verticals far less than angle does, and part of that is a data boundary rather than a market fact. Meta's own field distinguishes several near-identical carousel spellings, and our own write path records dynamic creative differently depending on which service wrote the document, so the line between a dynamic ad and a static one is partly a question of who ingested it. The video and static split is best read as a range rather than a point. The one format read that survives that caveat is the size of video. Nearly half of everything running in Malaysia is video, which is a much higher share than most published Malaysian creative advice assumes, and it holds across verticals rather than being driven by one or two.

Two limits belong on the page rather than in a footnote. The angle labels are assigned by a model reading the creative, not declared by the advertiser. They are a consistent reading rather than a truth, and a consistent reading is enough for an index, where both sides of the division carry the same bias, while being nowhere near enough for a decimal-point claim about a single ad. And the vocabulary drifts. The list of angles and formats in this repository is a copy of the classifier's, and copies rot: building this study found the format list carrying five values where the index emits ten, which had been silently dropping a whole carousel variant from every format cut, and the industry list missing a long tail of spelling variants of categories it already covers. Those variants carry a negligible number of ads between them, and every one of them ships in the CSV under its own dimension, so the drift is visible rather than assumed away.

The full matrix is one CSV under CC BY 4.0, free to reuse including commercially with attribution to AdPlay.ai. The page shows one row per vertical; the file carries every angle and every format for every vertical measured, each row naming whether its label came from Meta's own field or from a model, plus the coverage arithmetic and the vocabulary drift rows.